Tailoring Requirements Engineering for Responsible AI

Requirements Engineering is critical for AI accountability: The paper argues that traditional RE practices must be adapted specifically for AI systems, as standard approaches have proven insufficient—evidenced by documented acceptance issues with AI solutions in high-stakes domains like medicine, automotive, and scientific research after deployment.

Responsible AI requires capturing societal needs, not just technical specs: The paper emphasizes that RE for responsible AI must go beyond typical user and technical requirements to explicitly identify and document societal requirements, ensuring that broader ethical and social impacts are considered during system design rather than discovered as problems post-deployment.

Current RE practices create a gap in responsible AI development: The authors identify that existing RE disciplines, while useful for general software, are not adequately tailored for the unique challenges of AI systems, suggesting that organizations currently lack structured methodologies to systematically address responsible AI concerns during the requirements phase.

Early requirement definition prevents costly AI failures: By highlighting acceptance issues that emerge after deployment in critical domains, the paper demonstrates that inadequate RE for responsible AI leads to real-world harms and system rejections, implying that investing in better upfront requirements work could prevent these downstream problems.